abstractdog commented on a change in pull request #2225:
URL: https://github.com/apache/hive/pull/2225#discussion_r681820768
##########
File path:
ql/src/java/org/apache/hadoop/hive/ql/udf/ptf/ValueBoundaryScanner.java
##########
@@ -664,6 +629,84 @@ public Object computeValue(Object row) throws
HiveException {
*/
public abstract boolean isEqual(Object v1, Object v2);
+ protected int binarySearchBack(int rowId, PTFPartition p, Object sortKey,
int amt,
+ Order order) throws HiveException {
+ boolean isOrderDesc = order.equals(Order.DESC);
+ Object rowVal = sortKey;
+
+ int rMin = 0; // tracks lowest possible number fulfilling the range
requirement
+ int rMax = rowId; // tracks highest possible number fulfilling the range
requirement
+
+ boolean isDistanceGreater = true;
+ while (rMin < rMax) {
+ isDistanceGreater = isDistanceGreater(isOrderDesc ? rowVal : sortKey,
isOrderDesc ? sortKey : rowVal, amt);
+ if (isDistanceGreater) {
+ rMin = rowId + 1;
+ } else {
+ rMax = rowId;
+ }
+ rowId = rMin + (rMax - rMin) / 2;
+ rowVal = computeValueUseCache(rowId, p);
+ }
+ return rMin;
+ }
+
+ private int linearSearchBack(int r, PTFPartition p, Object sortKey, int amt,
+ Order order) throws HiveException {
+ boolean isOrderDesc = order.equals(Order.DESC);
+ Object rowVal = sortKey;
+ while (r >= 0 && !isDistanceGreater(isOrderDesc ? rowVal : sortKey,
+ isOrderDesc ? sortKey : rowVal, amt)) {
+ Pair<Integer, Object> stepResult = skipOrStepBack(r, p);
+ r = stepResult.getLeft();
+ rowVal = stepResult.getRight();
+ }
+ return r + 1;
+ }
+
+ protected int binarySearchForward(int rowId, PTFPartition p, Object sortKey,
int amt,
+ Order order) throws HiveException {
+ boolean isOrderDesc = order.equals(Order.DESC);
+ Object rowVal = sortKey;
+ int rMin = rowId; // tracks lowest possible number fulfilling the range
requirement
+ int rMax = p.size(); // tracks highest possible number fulfilling the
range requirement
+
+ boolean isDistanceGreater = true;
+ while (rMin < rMax) {
+ isDistanceGreater =
+ isDistanceGreater(isOrderDesc ? sortKey : rowVal, isOrderDesc ?
rowVal : sortKey, amt);
+ /*
+ * if the currently inspected row is the last (p.size() - 1) but we're
not far enough to fulfill
+ * the range requirement (!isDistanceGreater), we cannot move forward,
let's simply return p.size(),
+ * range end is exclusive
+ */
+ if (!isDistanceGreater && rowId == p.size() - 1) {
Review comment:
unfortunately, it's not the same, I tried it out and TestBoundaryCache
failed
in this case need to return p.size(), which makes sense when we're defining
a range boundary (this is what I meant with exclusive), but cannot use it for
indexing the records
when I changed initially rMax = p.size() - 1 and removed this short-circuit,
range calculation was incorrect (1 record smaller than expected, which is
because p.size() vs. p.size() -1)...if I just simply removed the short-circuit
(with having rMax = p.size()), I got IndexOutOfBoundsException
so this is the combination that works while binary searching to the end of a
partition
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
